Santiago A. Utsumi, PhD
Innovations for Sustainable
Farming & Ranching
Inspired by Nature, Driven by Science
The Digital Dairy Cow​
Dairy robotics, LoRa communication, GPS, embedded sensors and novel machine learning and deep learning tools are giving dairy farmers streams of Big data (with a capital "B") to digitize the phenotype of the modern dairy cow precisely, non-intrusively, and in real-time. A bright future unfolds to halt many of the ongoing global challenges dairy farmers are facing. For the experimental setup of the MSU Robot Dairy Farm, I calculate a fingerprint of over 10,000 data points, available daily to digitize each of the +140 milking cows. Similar to the "omics" technologies presently unfolding in many fields of sciences, dairy robotics could be seen as the emerging "multi-omic" platforms to digitize cows by unique differences in multiple behavior, metabolic and phenotypic traits. Just multiply the massive amount of cow data points by the potential number of cows in robot herds and you will get a realistic representation of the future for dairy management, in particular for scouting remotely indicators of cow health, comfort and well being, unique nutrition requirements, individualized feeding needs and breeding checks. I argue that the digital transformation for dairy management will most likely be hastened by the integration of emerging sources of data into the next generation of smart-cow monitoring systems. Here I draw on two relatively recent innovations undertaken at the lab.
Harnessing the Nutrition of Modern Robotic Cow​s
Progress for milk production is expected to shift from linear to exponential growth in the coming years. By 2050, the average milk yield for high producing cows (conventionally fed) is expected to reach 72 kg with ranges between 67 to 77 kg, as many experts suggest. There is no doubt that most of the predicted change will come from new "genomic selection" for milk and feed efficiency, yet this will demand for new on-farm approaches to cope increasingly with the changing requirements of modern dairy cows. Dairy robotics, machine learning, NDIR sensor, and acoustics will be among the new Smart-Feeding technologies.
Important breakthrough achievements have already been made. Smart-pulsation systems are allowing new releases of robots to milk cows 10 to 15% faster, so farmers may increase stocking rates by 15% or increase overall milk production up to a 24%. The Introduction of NDIR sensor in dairy robots enabled the real-time scouting of ruminant methane (CH4) and carbon dioxide (CO2) fluxes. This technology is offering a much better resolution of the dairy cow metabolic needs, nutrient use efficiency, homeorhesis, and metabolic health. Furthermore, CO2 respiration serves as metabolic bio-marker for dairy cows, and the non-invasive and automated quantification of CO2 through time allows making quick inferences on deviations for metabolic distress, changing levels of feed intake, or new estimates for Feed Efficiency based on residual differences of CO2 (or heat increment). Through smart-dairy gadgets nutritionists and dairy farmers will have new tools at hand to make smart-feeding decisions, better and faster, thereby nurturing closely the changing nature of modern dairy cows.
Decoding The Language of Feed Intake
Acoustic monitoring is among the most reliable methods to quantify biting and chewing, and a great deal of progress has been made on automation, and on estimation of feed intake using 'sound energy'. The technique is based on the discrimination of three distinct types of jaw movements: chews, bites, and the recently discovered composite chew-bites. In a second step, the temporal segmentation of jaw movement rates by a fixed time step is used to classify recordings into ingestive and rumination bouts.
One of the past limitations for applications of acoustics has been the exclusive focus on rumination, yet our most recent applications have successfully extend the analyses of sound signals for long-term monitoring of both rumination and ingestive chewing, faster and more precisely compared to presently widespread commercial systems. The present work is focused on deployments of improved embedded components for real-time and online detection and segmentation of biting and chewing, using improved machine learning tools. This technique will overcome past limitations for scalable applications of acoustics to large groups of cows that were related mainly to limited processing capabilities, power supply, and data storage capacity, and efficient data transfer.
Coupled NDIR sensors and air flow measurements in feed troughs of robotic milkers were installed for individualized quantification of CH4 and CO2 fluxes of dairy cows. Data was made available in real-time.
Above, I share the lactation profile for the CO2 flux, milk, dry matter intake (DMI), and body weight (BW) for the MSU robotic herd (142 cows; grey bars are 95% IC). Three peer review papers describe methods for CH4 and CO2 assessment in robots, and applications for determination of DMI, both for conventional and grazing. All data was automatically collected and processed daily.
Grazing is indivisibly link to the nature of ruminant herbivores, and novel acoustics allows better understanding of the regulation and controls of feed intake by grazing cattle. When confronted to a changing sward structure, chewing sounds unveil distinct arrays of jaw movements that are closely linked to a different regulation of intake rate. Consider the dairy cow as if she were a feed bin auger. This auger usually spins at constant rate, so the amount of feed taken by each auger spin varies with the characteristics and structure of that feed. The analogy to the example is that the 'appetite drive' makes dairy cows to consume feeds at nearly maximum jaw movement rates. As such, cows will increase intake rate by making more exclusive bites and chews when graze on tall usually more fibrous swards (left), or by reducing exclusive chews and bites and by increasing the number of composite chew-bites when graze on short or increasingly depleted swards that yield a lighter bite mass (right). HOW YOUR COWS SOUND? TOO MANY EXCLUSIVE CHEWS AND BITES? TOO MANY CHEW-BITES? Those are very likely questions that farmers will tweet in a near future. A bright application for acoustics unfolds to fine tune feeding management for a better control of feed intake and diets.
Drivers and determinants of feed intake:
Feed intake is multi-pronged,, emergent and defining trait, resulting from the interplay between an animal's requirements and the animal's drive for meeting and balancing out those requirements, behaviorally, nutritionally, physiologically and metabolically. Thus, feed intake links both mechanistic drivers and controls with stochastic factors, such as environmental constraints, For grazing dairy cattle feed intake introduces new dimensionalities and is indivisible connected to the ecology and evolutionary nature of grazing behavior. Grazing brings together many components, dimensions and controls by linking soils and plants, and animals and the atmosphere with people, and as such covaries along several biotic drivers, mechanistic controls, abiotic factors, and management behavior. For example, the amount of grass a dairy cow eats daily is the result of distinguishing foraging events aggregating hierarchically at defined domains of space and time.. Events, may include exclusive or overlapping scale-dependent behaviors such as food spatial search, food apprehension and gathering by lips and tongue, food chewing and food swallowing, all of which influence rates for herbage intake, digestion and nutrient utilization. Hypothetically, the voluntary food intake of dairy cows fed either standing forage crops or grassy swards, is mediated by plant traits and spatial heterogeneity, as animals graze down vegetation patches.
Here is how this feeding process game works. As dairy cows seek diets to meet requirements on bite by bite basis, the effects of bite mass and sward heterogeneity (i.e. variation of herbage mass in space) on intake rate are pivotal. All other factors steady, the slope for intake rate increase with increasing biting rates and bite mass, which in turn depends on changes in bite depth, bite area, and bite volume. The relationships between bite dimensions and bite mass with sward structure are usually positive. The most obvious functional links are: a) bites will be deeper and will have greater size and mass on taller and thicker swards, b) heavy bites (i.e. PMR cows) usually lead to more extensive ingestive chewing compared to bites of lower mass (i.e. grazing cows), yet, overlapping of bites and chews on same jaw movements, 'chew-biting', can serve as adaptation to buffer against declines of bite mass and intake rate, d) declining bite dimensions,, together with decreases in bite mass and intake rate occur as cows graze down swards through depleting horizons. This mechanism is a defining factor for changes in grazing behavior and herbage intake rates with increasing grazing intensity or sward depletion rates. Back in the summer of 2001, I was confronted as young MS student with field experiments to verify this functional relationships and hypotheses (Right Figure). I admit this experience was so transformative that sparked on me a great degree of curiosity to continue investigating the "mysteries" of feed intake, till today, and as I share with you below.
If we were to scale-up, spatially and temporally, the above described behaviors, the decisions dairy cows would make regarding when to eat,, what to eat, and for how long to eat would involve sets of optimization rules, self-learning and coping styles. The literature suggest that dairy cows exposed to varying types and arrays of feeds can develop quickly (i.e. 5-6 days) adaptive strategies to cope readily and rapidly with environmental change. In the summer of 2011, I further tested this functional hypothesis with my fellow lab students and lab visiting URA and honors student Anais FAUCHILE (Purpan, Fr.), all assigned to complete grazing research at the Robotic Dairy Farm of Michigan State University. Using a well designed set of feeding trials, we exposed replicated groups of 3 dairy cows (n=4) on a training phase of 6 days followed by a 3 days phase for testing to forage on pasture arenas containing concentrate feed (red dots) offered either, as fixed feeding locations over time (red treatments) or as randomly placed feeding locations over time (green treatments). Behaviors including the extent of cow traffic, speed of movement concentrate selectivity, and cow traffic and search efficiency were recorded by GPS tracking and GIS analysis of cow movements and intake measurements .
Already wondering what happened? Results are shown in the figures at the right. Behaviors were markedly altered by the way feeds were offered, spatially and temporally. Cows trained to search preferred feed while feed was offered on a predictable pattern (red treatment) had lower speed of movement, traveled shorter distances, were more choose and efficient with regard to traffic and feeding as they were able to 'consume more of the preferred feed with the least travel effort'. Conversely, as cows were unable to develop a spatial reference for allocations with preferred feed (green treatment) they relied deeply on a rapidly developing working memory to cope increasingly with changing arrangements of feeds. These cows clearly developed certain degree of 'behavioral anxiety' for consuming concentrate feed by traveling faster and by exploring the entire feeding arena to a greater extent, but at greater search costs and lower selectivity, thereby penalizing feed efficiency search, cow traffic and feed intake (green bars). Thus, cow traffic, intake and milking of robotic dairy cows would be altered and improved depending on feeding management. Whether robotic dairy cows are confronted with fairly predictable and consistent feeding schedules (enabling reference memory) or exposed to inconsistent feeding sequences (forcing greater reliance on short-term working memory), can determine largely the success and ability to entice patterns of desirable cow traffic in robotic dairy herds, and regardless of whether cows are fed conventionally or on grass-based diets.
Precision feeding management is a timely and increasingly relevant issue to robotic 'cow traffic' management. The issue is so defining that much of the success of managing robot cow will relay on our ability to alter non-intrusively many of the decisions dairy cows would make regarding when and how to eat, Enticing feeding patterns based either on developments of reference vs. working memory could be relevant, in particular for grazing dairy cows increasingly exposed to challenging feeding schedules (i.e. greater walking distance to resources, climate factors. low intake rate, etc.). A clear understanding of how feed allocations will influence rates of herbage depletion, bite mass and slopes for herbage intake rate is critical. To facilitate exploring those effects I facilitate 'SmartGrassIntake', a grazing simulation tool that you can use to explore feeding patterns and behaviors for your own cows and as affected by the chemistry, height, herbage mass and distribution of herbage mass for your own pastures . Just let me know how the exercise goes.
​